I have NGINX setup with PHP7.2-FPM and am having issues with burst requests. The average amount of active processes (fpm children) is 15.
I have these settings (48 core server):
listen.backlog = 1024
pm.max_children = 1024
pm.start_servers = 16
pm.min_spare_servers = 8
pm.max_spare_servers = 16
pm.max_requests = 5000
I also tested a min. amount of spare servers of 256 which lowered but did not resolve the amount of Resource temporarily unavailable
. Obviously this is not a setting for production and was merely done to see if this many workers could deal with the size of bursts.
I also tried a listen.backlog of 65536 which didn't make a difference.
What is the way to go? If it would queue the requests there not be any issues.
Edit: I just noticed that fpm-status always reports max listen queue
as 0